Speech is an online voice recorder that checks and helps you improve your pronunciation.
Try it online at: https://speech.sege.dev
As a non-native English speaker, I struggle with my pronunciation a lot. This is a tool I wished I had but could not find it anywhere, so I decided to build it myself.
Or maybe it's just a good excuse for yet another side project.
Under the hood, Speech uses Vosk – the speech recognition toolkit, to check your voice recording and figure out what you are trying to say.
The audio recording is done using the Web Audio API, and encoded into WAV format using the WebAudioRecorder.js library.
First, you need to download a Vosk model at https://alphacephei.com/vosk/models and extract it to the model folder in the source directory.
Also, you will need to have libvosk.so or libvosk.dylib in your computer's library path. You can download it from the vosk-api repository, and copy the library file to somewhere like /usr/local/lib.
Then you are ready to run the project:
$ cargo run
The frontend will be built automatically using Parcel, and the server will be started at http://localhost:3000.
The speech score provided by this tool is based on the confidence score of each word provided by Vosk's model.
The pronunciation data is provided by The CMU Pronouncing Dictionary.
The speech examples are collected from the Random Sentence Generator website.